eigen() strikes again

[Ilya Zakharevich <ilya@math.ohio-state.edu>]

With a week-old version:

    ? mateigen([1,2;3,4])
      ***   precision too low in eigen.
    ? \p 38 
       realprecision = 38 significant digits
    ? mateigen([1,2;3,4])
      ***   precision too low in eigen.
    ? \p80
       realprecision = 86 significant digits (80 digits displayed)
    ? mateigen([1,2;3,4])
    %1 = 
    [-1.45742710...

Looking at the code for the Gauss algorithm (in 2.0.15, which would
just silently give a wrong answer), I would expect that the correct
message should be not "precision too low", but "some random collection
of coincidences did not happen"...  Right?

Ilya